Help creating a LUT: BT709 to PQ

[FranceBB]

Hi,
I have a very particular request.
I've got the raw files of an interview of a famous football player. The interview has been shot in 3840x2160 1.77 FF 10bit BT2020 HDR, 4:2:0 Type 2 with the following specs:

Standard : Component
Color space : YUV
Chroma subsampling : 4:2:0 (Type 2)
Bit depth : 10 bits
Color range : Limited
Color primaries : BT.2020
Transfer characteristics : SMPTE ST 2084
Matrix coefficients : BT.2020 non-constant
Mastering display color primaries : R: x=0.680000 y=0.320000, G: x=0.265000 y=0.690000, B: x=0.150000 y=0.060000, White point: x=0.312700 y=0.329000
Mastering display luminance : min: 0.0000 cd/m2, max: 1000.0000 cd/m2
Maximum Content Light Level : 1000 cd/m2
Maximum Frame-Average Light Level : 400 cd/m2


So far so good.
Here comes the tricky part.
While the player is talking, the interviewer is gonna show some footages of his old games. Some of them are old XDCAM50 in BT709 and some other were old BetaCAM that I had to acquire from scratch. My target file is PQ, in other words BT2020 HDR10 and the specs must match the ones of the HDR file provided, which means that I gotta:

1) Deinterlace
tdeint(mode=2, order=-1, field=-1, mthreshL=6, mthreshC=6, map=0, type=2, debug=false, mtnmode=1, sharp=true, cthresh=6, blockx=16, blocky=16, chroma=true, MI=64, tryWeave=true, link=1, denoise=true, slow=2, opt=4)

2) Upscale to 3840x2160
nnedi3_resize16(target_width=1920, target_height=1080, mixed=true, thr=1.0, elast=1.5, nns=4, qual=2, etype=0, pscrn=4, threads=0, tv_range=true, kernel_d="Spline", kernel_u="Spline", taps=6, f_d=1.0, f_u=2.0, sharp=0, lsb_in=true, lsb=true)

3) Deband and convert 16bit stacked to 16bit interleaved
flash3kyuu_deband_f3kdb(range=15, Y=64, Cb=64, Cr=64, grainY=64, grainC=64, sample_mode=2, blur_first=True, dynamic_grain=False, opt=-1, mt=True, dither_algo=3, keep_tv_range=False, input_mode=1, input_depth=16, output_mode=2, output_depth=16)

4) Convert from 16bit interleaved to real 16bit
avs2yuv.exe "H:\AVS Script.avs" -depth 16 -csp auto - |

5) Convert from BT709 to BT2020 with 16bit precision and 4:2:0 to 4:2:0 Type2
ffmpeg.exe -i - -vf lut3d="Custom_LUT_2.cube" -pix_fmt yuv420p16le -f yuv4mpegpipe -strict -1 - | ffmpeg.exe -i - -vf scale=out_color_matrix=bt2020nc:out_h_chr_pos=0:out_v_chr_pos=0 -pix_fmt yuv420p16le -strict -1 -an -f yuv4mpegpipe - | x265.exe --y4m - --dither


6) Convert from BT2020 4:2:0 Type2 SDR to PQ / HDR10
-vf lut3d="Custom_LUT.cube"

(The red part is the one I don't know how to do).

I was thinking about creating a LUT, but I've never done anything like this before... I mean, I never graded an SDR to HDR10 (PQ). Looking for solutions on google I found this site: https://www.mysterybox.us/store/sdr-to-hdr10-conversionsv10 but I would rather not spend 200 bucks for that.

Any thoughts? Thank you in advance.

[Blue_MiSfit]

Maybe use Resolve or some other NLE for this? I think they'll probably do it pretty well.

[videoh]

Does it have to be a LUT? I could crank out a CUDA Avisynth+ filter (100+ fps for 3840x2160) in no time.

[FranceBB]

Does it have to be a LUT? I could crank out a CUDA Avisynth+ filter (100+ fps for 3840x2160) in no time.

It doesn't have to be a LUT. I was thinking about a LUT 'cause everybody seems to be using LUTs for this kind of things.

I ended up making one and I tweaked levels and so on, but it I can't get colors right, especially red.

This is the LUT I created. Very "light" chroma, but still... https://mega.nz/#!3N0TSSLT!KkhslQPudhXBZMW0yKBGzhwm6lS5AYxUcIV2BZzxuw0

[Cary Knoop]

What camera shoots directly PQ?
Are you sure these are the raw files, seems more like graded material.

I would use Resolve (or any other tool that does color space transforms) take Rec2100 ST2084 as the input colorspace, use your own timeline colorspace and make the output colorspace Rec.709.
Obviously you would have to map the wider space onto the smaller Rec709 gamma and you also may have to do some gamut mapping to the smaller Rec.709 gamut.

[FranceBB]

Are you sure these are the raw files, seems more like graded material.

They are graded, but we relied on an external service to record the interview and that's what they sent us. Not really sure what they did, but what they sent us is our "starting file" (that's what I meant with "raw").

[FranceBB]

For everybody reading (in the future as well) I solved the problem by creating my own LUTs.
In order to convert Rec709 to Rec2020 PQ, I tried to scale the Rec709 curve "as it is", but it ended up being too dark (of course).
Then I came across this White Paper by BBC https://pdfs.semanticscholar.org/daf8/b3789fb19961d604c087ddc48f0884f61cd1.pdf and it opened my mind.
So, I made two LUTs, with Rec709 scaled by 400% and 800% and this somehow worked out as expected with results really close to the original Rec709 version and only slightly darker, but in PQ Rec2020:

BT709_to_PQ_400: https://mega.nz/#!3N0TSSLT!KkhslQPudhXBZMW0yKBGzhwm6lS5AYxUcIV2BZzxuw0
BT709_to_PQ_800: https://mega.nz/#!DdMhUCgC!djQpewXxZr8YXQoeBhz-KnBRAhnZ-T7FbHnVoYKV2l0

Please note that I limited the output to 1000nit.
Once again, guys, please note that the intent of these two LUTs is to make Rec709 SDR look almost identical in Rec2020 PQ HDR 'cause I'm gonna encode everything in Rec2020 PQ HDR due to the fact that I have mixed contents (SDR and HDR).

[Cary Knoop]

You can't make footage that is recorded in Rec.709 into HDR. It could be possible to make footage that is recorded in high dynamic range and graded in 10 bit (or higher) Rec709 turn into HDR but it depends.
And obviously, while you can put a Rec.709 color space into a Rec.2020 container you cannot invent colors beyond the Rec709 gamut.

[FranceBB]

@Cary Knoop... of course, making Rec.709 in PQ won't make it look better; what I needed was different. I had mixed sources: a legacy SDR with Rec709 and an HDR with Rec2020 PQ.
I needed to encode them together in a single video file as Rec2020 PQ, so I had to scale the SDR Rec709 parts to be displayed correctly in Rec2020 color space using the PQ curve.
This way, they are gonna look exactly identical to the original Rec.709 SDR file, but they will be encoded in a Rec2020 PQ file.
In other words, I can encode the scenes of the journalist interviewing the player (real HDR Rec2020 PQ), and the footages of the old games that were in SDR Rec709 all together in Rec2020 PQ.
Scaling them is not gonna add any additional information, but the "goal" is to encode legacy contents together with new contents in an HDR Rec2020 PQ file. ;)

[Cary Knoop]

@Cary Knoop... of course, making Rec.709 in PQ won't make it look better; what I needed was different. I had mixed sources: a legacy SDR with Rec709 and an HDR with Rec2020 PQ.
I needed to encode them together in a single video file as Rec2020 PQ, so I had to scale the SDR Rec709 parts to be displayed correctly in Rec2020 color space using the PQ curve.
This way, they are gonna look exactly identical to the original Rec.709 SDR file, but they will be encoded in a Rec2020 PQ file.
In other words, I can encode the scenes of the journalist interviewing the player (real HDR Rec2020 PQ), and the footages of the old games that were in SDR Rec709 all together in Rec2020 PQ.
Scaling them is not gonna add any additional information, but the "goal" is to encode legacy contents together with new contents in an HDR Rec2020 PQ file. ;)
You are absolutely right with respect to the need for mapping the Rec709 to the Rec2020 gamut otherwise the colors will become oversaturated.

With respect to mapping the gamma curves what diffused white point do you use for the Rec709 mapped onto PQ and how do you treat the blacks?

[FranceBB]

Nothing particularly complicated, really.
On the left there's the graph representing the LUT I used (thanks to the BBC White Paper I linked above) and on the right there's the official ITU Proposal:

https://i.imgur.com/YFjToim.png

Red: Rec709 scaled to 800% (BBC White Paper - ITU Proposal).
Blue: PQ Curve (limited to 1000nit).

I didn't specify a "knee", otherwise it would have been different.

[zn]

For everybody reading (in the future as well) I solved the problem by creating my own LUTs.
In order to convert Rec709 to Rec2020 PQ, I tried to scale the Rec709 curve "as it is", but it ended up being too dark (of course).
Then I came across this White Paper by BBC https://pdfs.semanticscholar.org/daf8/b3789fb19961d604c087ddc48f0884f61cd1.pdf and it opened my mind.
So, I made two LUTs, with Rec709 scaled by 400% and 800% and this somehow worked out as expected with results really close to the original Rec709 version and only slightly darker, but in PQ Rec2020:

BT709_to_PQ_400:
BT709_to_PQ_800:

Please note that I limited the output to 1000nit.
Once again, guys, please note that the intent of these two LUTs is to make Rec709 SDR look almost identical in Rec2020 PQ HDR 'cause I'm gonna encode everything in Rec2020 PQ HDR due to the fact that I have mixed contents (SDR and HDR).

Hi. can you re-upload lut files, please?

[FranceBB]

Hi. can you re-upload lut files, please?

They're all here: https://github.com/FranceBB/LinearTransformation

Including the "BT709_to_PQ.cube"

[wswartzendruber]

BT.2446 covers displaying SDR within HDR. It presents something like three methods.

EDIT: The jury seems to be out as to whether or not OOTF gamma adjustment needs to be done when scaling linear display light. BT.2446 says yes for some methods. The principle of BT.1886 says no.

EDIT: BT.2446 also talks about going from linear display light to XYZ, and then to CIE 1976 Yu'v'. The point is to scale Y so that chromacity is unaffected. I have found that simply scaling display linear RGB produces mathematically identical results.